.. index:: 
     single: R

.. _mdoc_R:

===========
R
===========

R is available for statistical computing. There are R modules built with both
the GCC and Intel compilers. We recommend using the Intel builds since those
have had the best performance during benchmarks. Some R packages may not
compile correctly with the Intel compilers, so use the GCC version in that
case. All R modules have been built with OpenMP enabled and use the Intel MKL
to improve performance. 

The currently available R modules::

    $ module avail R
    ------------------------- /software/modulefiles -----------------------------
    R/2.15(default)   R/2.15+intel-12.1 R/3.0             R/3.0+intel-12.1  R/3.1             R/3.1+intel-15.0

To install and use additional R packages to your home directory, it is
necessary to set to set the environment variable ``R_LIBS_USER``. For example::

    export R_LIBS_USER=$HOME/R_libs

The directory specified should exist before trying to install R packages.

The RStudio IDE is also available as the ``rstudio`` module. This provides
a graphical interface for developing and running R. To use R in this mode,
you should connect to midway via :ref:`thinlinc`.

.. contents:: :local:

Serial Examples
===============

Here is a simple "hello world" example to submit an R job to Slurm. This is
appropriate for an R job that expects to use a single CPU.

sbatch script :download:`Rhello.sbatch`

.. literalinclude:: Rhello.sbatch
        :language: bash

R script :download:`Rhello.R`:

.. literalinclude:: Rhello.R

Output:

.. literalinclude:: Rhello.out


Parallel Examples
=================

For parallel use there are several options depending on whether there should be
parallel tasks on a single node only or multiple nodes and the level of
flexibility required.  There are other R packages available for parallel
programming than what is covered here, but we'll cover some frequently used
packages.

Multicore
---------

On a single node, it is possible to use doParallel and foreach.

sbatch script :download:`doParallel.sbatch`:

.. literalinclude:: doParallel.sbatch
        :language: bash

R script :download:`doParallel.R`:

.. literalinclude:: doParallel.R

Output:

.. literalinclude:: doParallel.out

SNOW
----

For multiple nodes, you can use the SNOW package, which provides a select
number of functions to simplify using multi-node clusters.

sbatch script :download:`snow-test.sbatch`:

.. literalinclude:: snow-test.sbatch
        :language: bash

R script :download:`snow-test.R`:

.. literalinclude:: snow-test.R

Output (trimmed for readability):

.. literalinclude:: snow-test.out

Rmpi
----

For multiple nodes, you can also use Rmpi. This is what snow uses internally.
It is less convenient than snow, but also more flexible.

This page has a number of useful Rmpi examples: http://www.umbc.edu/hpcf/resources-tara-2010/how-to-run-R.php

sbatch script :download:`Rmpi.sbatch`:

.. literalinclude:: Rmpi.sbatch
        :language: bash

R script :download:`Rmpi.R`:

.. literalinclude:: Rmpi.R

Output (trimmed for readability):

.. literalinclude:: Rmpi.out